Post on 11-Jul-2020
1/12
JI
Back
FullScr
New quasi-symmetric designs by the
Kramer-Mesner method?
Vedran Krcadinac Renata Vlahovic
University of Zagreb, Croatia and University of Zagreb, Croatia
vedran.krcadinac@math.hr renata.vlahovic@math.hr
? Supported in part by the Croatian Science Foundation under project 1637 and by the European Science Foundationunder COST Action IC1104.
2/12
JI
Back
FullScr
A t-(v, k, λ) design is a set V = {1, . . . , v} of points together with afamily B = {B1, . . . , Bb} of k-element subsets of V called blocks suchthat any t-subset of points is contained in exactly λ blocks.
2/12
JI
Back
FullScr
A t-(v, k, λ) design is a set V = {1, . . . , v} of points together with afamily B = {B1, . . . , Bb} of k-element subsets of V called blocks suchthat any t-subset of points is contained in exactly λ blocks.
The total number of blocks b and the number of blocks r through anygiven point can be computed from t, v, k and λ.
b = λ ·(vt
)(kt
), r = λ ·(v−1t−1
)(k−1t−1
)
2/12
JI
Back
FullScr
A t-(v, k, λ) design is a set V = {1, . . . , v} of points together with afamily B = {B1, . . . , Bb} of k-element subsets of V called blocks suchthat any t-subset of points is contained in exactly λ blocks.
The total number of blocks b and the number of blocks r through anygiven point can be computed from t, v, k and λ.
b = λ ·(vt
)(kt
), r = λ ·(v−1t−1
)(k−1t−1
)An automorphism of the design is a permutation α : V → V takingblocks to blocks, i.e. such that α(B) ∈ B for any B ∈ B.
3/12
JI
Back
FullScr
Let G be a group of permutations of V and let T1, . . . , Tm be theorbits of t-element subsets of V and K1, . . . ,Kn the orbits of k-elementsubsets of V .
3/12
JI
Back
FullScr
Let G be a group of permutations of V and let T1, . . . , Tm be theorbits of t-element subsets of V and K1, . . . ,Kn the orbits of k-elementsubsets of V .
Let aij = # of elements of Kj containing a given T ∈ Ti(does not depend on the choice of T ).
3/12
JI
Back
FullScr
Let G be a group of permutations of V and let T1, . . . , Tm be theorbits of t-element subsets of V and K1, . . . ,Kn the orbits of k-elementsubsets of V .
Let aij = # of elements of Kj containing a given T ∈ Ti(does not depend on the choice of T ).
The matrix A = [aij] is the Kramer-Mesner matrix. Designs with G asan automorphism group correspond to 0–1 solutions of the system oflinear equations A · x = λ J , where J = (1, . . . , 1)τ .
3/12
JI
Back
FullScr
Let G be a group of permutations of V and let T1, . . . , Tm be theorbits of t-element subsets of V and K1, . . . ,Kn the orbits of k-elementsubsets of V .
Let aij = # of elements of Kj containing a given T ∈ Ti(does not depend on the choice of T ).
The matrix A = [aij] is the Kramer-Mesner matrix. Designs with G asan automorphism group correspond to 0–1 solutions of the system oflinear equations A · x = λ J , where J = (1, . . . , 1)τ .
Solving systems of linear equations over the integers is a NP completeproblem. The Kramer-Mesner system is computationally feasible only ifthe number of variables n is sufficiently small.
4/12
JI
Back
FullScr
A t-(v, k, λ) design is quasi-symmetric if any two blocks intersect eitherin x or in y points, for non-negative integers x < y. Quasi-symmetricdesigns have important connections with strongly regular graphs andself-orthogonal codes.
4/12
JI
Back
FullScr
A t-(v, k, λ) design is quasi-symmetric if any two blocks intersect eitherin x or in y points, for non-negative integers x < y. Quasi-symmetricdesigns have important connections with strongly regular graphs andself-orthogonal codes.
It is known that quasi-symmetric designs:
• do not exist for t ≥ 5;
4/12
JI
Back
FullScr
A t-(v, k, λ) design is quasi-symmetric if any two blocks intersect eitherin x or in y points, for non-negative integers x < y. Quasi-symmetricdesigns have important connections with strongly regular graphs andself-orthogonal codes.
It is known that quasi-symmetric designs:
• do not exist for t ≥ 5;
• for t = 4 the only example is the 4-(23, 7, 1) design (x = 1, y = 3)and its complement;
4/12
JI
Back
FullScr
A t-(v, k, λ) design is quasi-symmetric if any two blocks intersect eitherin x or in y points, for non-negative integers x < y. Quasi-symmetricdesigns have important connections with strongly regular graphs andself-orthogonal codes.
It is known that quasi-symmetric designs:
• do not exist for t ≥ 5;
• for t = 4 the only example is the 4-(23, 7, 1) design (x = 1, y = 3)and its complement;
• for t = 3 it is conjectured that the only examples are
– the quasi-symmetric 4-design and its residual 3-(22, 7, 4) design,
– Hadamard 3-designs,
– 3-((λ + 1)(λ2 + 5λ + 5), (λ + 1)(λ + 2), λ) designs(known to exist only for λ = 1),
– a hypothetical 3-(496, 40, 3) design,
– complements of these designs.
5/12
JI
Back
FullScr
For t = 2 the classification of feasible parameters of quasi-symmetricdesigns is widely open. There are many parameter sets for which exis-tence is unknown.
5/12
JI
Back
FullScr
For t = 2 the classification of feasible parameters of quasi-symmetricdesigns is widely open. There are many parameter sets for which exis-tence is unknown.
M. S. Shrikhande, Quasi-symmetric designs, in: The Handbook of Com-binatorial Designs, Second Edition (eds. C. J. Colbourn and J. H. Dinitz),CRC Press, 2007, pp. 578–582.
5/12
JI
Back
FullScr
For t = 2 the classification of feasible parameters of quasi-symmetricdesigns is widely open. There are many parameter sets for which exis-tence is unknown.
M. S. Shrikhande, Quasi-symmetric designs, in: The Handbook of Com-binatorial Designs, Second Edition (eds. C. J. Colbourn and J. H. Dinitz),CRC Press, 2007, pp. 578–582.
6/12
JI
Back
FullScr
Adaptations of Kramer-Mesner method to quasi-symmetric designs:
1. An orbit of k-subsets Ki is good if |K1 ∩ K2| is either x or y, forany two elements K1, K2 ∈ Ki. We can limit the search to good orbitsand thereby reduce the number of variables n.
6/12
JI
Back
FullScr
Adaptations of Kramer-Mesner method to quasi-symmetric designs:
1. An orbit of k-subsets Ki is good if |K1 ∩ K2| is either x or y, forany two elements K1, K2 ∈ Ki. We can limit the search to good orbitsand thereby reduce the number of variables n.
2. Two orbits Ki, Kj are compatible if |K1 ∩K2| is either x or y, forany K1 ∈ Ki, K2 ∈ Kj. The n × n matrix C = [cij] with cij = 1 ifKi and Kj are compatible, and cij = 0 otherwise, is the compatibilitymatrix. This information can be used to make the backtracking searchfor solutions of the Kramer-Mesner system more efficient.
7/12
JI
Back
FullScr
Example: 2-(28, 12, 11), x = 4, y = 6. Group G = 〈α, β〉 isomorphicto the dihedral group of order 12 generated by the permutations
α = (1, 2, 3, 4, 5, 6)(7, 8, 9, 10, 11, 12)(13, 14, 15, 16, 17, 18)(19, 20, 21, 22, 23, 24)(25, 26, 27),
β = (1, 6)(2, 5)(3, 4)(7, 11)(8, 10)(13, 17)(14, 16)(19, 23)(20, 22)(25, 27).
7/12
JI
Back
FullScr
Example: 2-(28, 12, 11), x = 4, y = 6. Group G = 〈α, β〉 isomorphicto the dihedral group of order 12 generated by the permutations
α = (1, 2, 3, 4, 5, 6)(7, 8, 9, 10, 11, 12)(13, 14, 15, 16, 17, 18)(19, 20, 21, 22, 23, 24)(25, 26, 27),
β = (1, 6)(2, 5)(3, 4)(7, 11)(8, 10)(13, 17)(14, 16)(19, 23)(20, 22)(25, 27).
V = {1, . . . , 28}Number of orbits T1, . . . , Tm of 2-element subsets: m = 47.
7/12
JI
Back
FullScr
Example: 2-(28, 12, 11), x = 4, y = 6. Group G = 〈α, β〉 isomorphicto the dihedral group of order 12 generated by the permutations
α = (1, 2, 3, 4, 5, 6)(7, 8, 9, 10, 11, 12)(13, 14, 15, 16, 17, 18)(19, 20, 21, 22, 23, 24)(25, 26, 27),
β = (1, 6)(2, 5)(3, 4)(7, 11)(8, 10)(13, 17)(14, 16)(19, 23)(20, 22)(25, 27).
V = {1, . . . , 28}Number of orbits T1, . . . , Tm of 2-element subsets: m = 47.
Total number of orbits of 12-element subsets: n = 2 543 568.
7/12
JI
Back
FullScr
Example: 2-(28, 12, 11), x = 4, y = 6. Group G = 〈α, β〉 isomorphicto the dihedral group of order 12 generated by the permutations
α = (1, 2, 3, 4, 5, 6)(7, 8, 9, 10, 11, 12)(13, 14, 15, 16, 17, 18)(19, 20, 21, 22, 23, 24)(25, 26, 27),
β = (1, 6)(2, 5)(3, 4)(7, 11)(8, 10)(13, 17)(14, 16)(19, 23)(20, 22)(25, 27).
V = {1, . . . , 28}Number of orbits T1, . . . , Tm of 2-element subsets: m = 47.
Total number of orbits of 12-element subsets: n = 2 543 568.
Number of good orbits K1, . . . ,Kn: n = 1097.
7/12
JI
Back
FullScr
Example: 2-(28, 12, 11), x = 4, y = 6. Group G = 〈α, β〉 isomorphicto the dihedral group of order 12 generated by the permutations
α = (1, 2, 3, 4, 5, 6)(7, 8, 9, 10, 11, 12)(13, 14, 15, 16, 17, 18)(19, 20, 21, 22, 23, 24)(25, 26, 27),
β = (1, 6)(2, 5)(3, 4)(7, 11)(8, 10)(13, 17)(14, 16)(19, 23)(20, 22)(25, 27).
V = {1, . . . , 28}Number of orbits T1, . . . , Tm of 2-element subsets: m = 47.
Total number of orbits of 12-element subsets: n = 2 543 568.
Number of good orbits K1, . . . ,Kn: n = 1097.
Number of solutions of the 47×1097 Kramer-Mesner system respectingthe compatibility matrix: 654 336.
7/12
JI
Back
FullScr
Example: 2-(28, 12, 11), x = 4, y = 6. Group G = 〈α, β〉 isomorphicto the dihedral group of order 12 generated by the permutations
α = (1, 2, 3, 4, 5, 6)(7, 8, 9, 10, 11, 12)(13, 14, 15, 16, 17, 18)(19, 20, 21, 22, 23, 24)(25, 26, 27),
β = (1, 6)(2, 5)(3, 4)(7, 11)(8, 10)(13, 17)(14, 16)(19, 23)(20, 22)(25, 27).
V = {1, . . . , 28}Number of orbits T1, . . . , Tm of 2-element subsets: m = 47.
Total number of orbits of 12-element subsets: n = 2 543 568.
Number of good orbits K1, . . . ,Kn: n = 1097.
Number of solutions of the 47×1097 Kramer-Mesner system respectingthe compatibility matrix: 654 336.
Number of non-isomorphic designs: 13 656.
7/12
JI
Back
FullScr
Example: 2-(28, 12, 11), x = 4, y = 6. Group G = 〈α, β〉 isomorphicto the dihedral group of order 12 generated by the permutations
α = (1, 2, 3, 4, 5, 6)(7, 8, 9, 10, 11, 12)(13, 14, 15, 16, 17, 18)(19, 20, 21, 22, 23, 24)(25, 26, 27),
β = (1, 6)(2, 5)(3, 4)(7, 11)(8, 10)(13, 17)(14, 16)(19, 23)(20, 22)(25, 27).
V = {1, . . . , 28}Number of orbits T1, . . . , Tm of 2-element subsets: m = 47.
Total number of orbits of 12-element subsets: n = 2 543 568.
Number of good orbits K1, . . . ,Kn: n = 1097.
Number of solutions of the 47×1097 Kramer-Mesner system respectingthe compatibility matrix: 654 336.
Number of non-isomorphic designs: 13 656.
We use GAP to compute the orbits and set up the Kramer-Mesnersystem, our own backtracking solver written in C, and nauty2 byB. D. McKay and A. Piperno for isomorphism checking.
8/12
JI
Back
FullScr
Proposition. Up to isomorphism there are 13656 quasi-symmetric(28, 12, 11) designs with x = 4, y = 6 and G = 〈α, β〉 as an auto-morphism group.
8/12
JI
Back
FullScr
Proposition. Up to isomorphism there are 13656 quasi-symmetric(28, 12, 11) designs with x = 4, y = 6 and G = 〈α, β〉 as an auto-morphism group.
Example: 2-(36, 16, 12), x = 6, y = 8. Group G = 〈α, β〉 isomorphicto the symmetric group S4 generated by the permutations
α = (1, 2, 3)(4, 5, 6)(7, 8, 9)(10, 11, 12)(13, 14, 15)(16, 17, 18)(19, 20, 21)(22, 23, 24)(25, 26, 27)(28, 29, 30)(31, 32, 33),
β = (1, 4)(2, 7)(5, 9)(6, 11)(8, 10)(13, 16)(14, 19)(15, 21)(17, 22)(18, 24)(20, 23)(26, 27)(29, 30)(32, 34).
8/12
JI
Back
FullScr
Proposition. Up to isomorphism there are 13656 quasi-symmetric(28, 12, 11) designs with x = 4, y = 6 and G = 〈α, β〉 as an auto-morphism group.
Example: 2-(36, 16, 12), x = 6, y = 8. Group G = 〈α, β〉 isomorphicto the symmetric group S4 generated by the permutations
α = (1, 2, 3)(4, 5, 6)(7, 8, 9)(10, 11, 12)(13, 14, 15)(16, 17, 18)(19, 20, 21)(22, 23, 24)(25, 26, 27)(28, 29, 30)(31, 32, 33),
β = (1, 4)(2, 7)(5, 9)(6, 11)(8, 10)(13, 16)(14, 19)(15, 21)(17, 22)(18, 24)(20, 23)(26, 27)(29, 30)(32, 34).
Proposition. Up to isomorphism there are 35572 quasi-symmetric(36, 16, 12) designs with x = 6, y = 8 and G = 〈α, β〉 as an auto-morphism group.
8/12
JI
Back
FullScr
Proposition. Up to isomorphism there are 13656 quasi-symmetric(28, 12, 11) designs with x = 4, y = 6 and G = 〈α, β〉 as an auto-morphism group.
Example: 2-(36, 16, 12), x = 6, y = 8. Group G = 〈α, β〉 isomorphicto the symmetric group S4 generated by the permutations
α = (1, 2, 3)(4, 5, 6)(7, 8, 9)(10, 11, 12)(13, 14, 15)(16, 17, 18)(19, 20, 21)(22, 23, 24)(25, 26, 27)(28, 29, 30)(31, 32, 33),
β = (1, 4)(2, 7)(5, 9)(6, 11)(8, 10)(13, 16)(14, 19)(15, 21)(17, 22)(18, 24)(20, 23)(26, 27)(29, 30)(32, 34).
Proposition. Up to isomorphism there are 35572 quasi-symmetric(36, 16, 12) designs with x = 6, y = 8 and G = 〈α, β〉 as an auto-morphism group.
Theorem. There are more than 50 000 quasi-symmetric (28, 12, 11)designs and more than 500 000 quasi-symmetric (36, 16, 12) designs upto isomorphism.
9/12
JI
Back
FullScr
Example: 2-(56, 16, 18), x = 4, y = 8 (unknown).
How to choose the group G?
9/12
JI
Back
FullScr
Example: 2-(56, 16, 18), x = 4, y = 8 (unknown).
How to choose the group G?
The associated strongly regular graph has parameters SRG(231, 30, 9, 3).Such a graph exists (the Cameron graph) and has a group of automor-phisms isomorphic to the Mathieu group M21 of order 20160.
9/12
JI
Back
FullScr
Example: 2-(56, 16, 18), x = 4, y = 8 (unknown).
How to choose the group G?
The associated strongly regular graph has parameters SRG(231, 30, 9, 3).Such a graph exists (the Cameron graph) and has a group of automor-phisms isomorphic to the Mathieu group M21 of order 20160.
The group M21 also acts on the quasi-symmetric 2-(21, 6, 4), x = 0,y = 2 design with b = 56 blocks. It has a subgroup G of order 960isomorphic to (Z2 × Z2 × Z2 × Z2).A5. We take the permutationrepresentation of degree 56 determined by the action on the blocks ofthe (21, 6, 4) design: G = 〈α, β〉,
α = (1, 2, 3, 4, 5)(6, 7, 8, 9, 10)(11, 12, 13, 14, 15)(16, 17, 18, 19, 20)(21, 22, 23, 24, 25)(26, 27, 28, 29, 30)(31, 32, 33, 34, 35)(36, 37, 38, 39, 40)(41, 42, 43, 44, 45)(46, 47, 48, 49, 50)(51, 52, 53, 54, 55),
β = (1, 6, 8)(2, 21, 26)(3, 32, 34)(4, 11, 5)(7, 15, 22)(9, 16, 13)(10, 29, 17)(12, 33, 30)(14, 19, 31)(18, 23, 35)(24, 28, 36)(25, 37, 39)(27, 38, 40)(42, 51, 49)(43, 52, 45)(44, 46, 47)(48, 54, 53)(50, 56, 55).
10/12
JI
Back
FullScr
Theorem. There are three quasi-symmetric (56, 16, 18) designs withx = 4, y = 8 and G = 〈α, β〉 as an automorphism group. The firstone has G as its full automorphism group, the second one has a splitextension G.Z2 of order 1920, and the third one has a split extensionM21.Z2 of order 40320 as its full automorphism group.
10/12
JI
Back
FullScr
Theorem. There are three quasi-symmetric (56, 16, 18) designs withx = 4, y = 8 and G = 〈α, β〉 as an automorphism group. The firstone has G as its full automorphism group, the second one has a splitextension G.Z2 of order 1920, and the third one has a split extensionM21.Z2 of order 40320 as its full automorphism group.
Computational details:
• total number of orbits of 16-element subsets is >(56
16)960 ≈ 4.34 · 1010
10/12
JI
Back
FullScr
Theorem. There are three quasi-symmetric (56, 16, 18) designs withx = 4, y = 8 and G = 〈α, β〉 as an automorphism group. The firstone has G as its full automorphism group, the second one has a splitextension G.Z2 of order 1920, and the third one has a split extensionM21.Z2 of order 40320 as its full automorphism group.
Computational details:
• total number of orbits of 16-element subsets is >(56
16)960 ≈ 4.34 · 1010
• most orbits are of size 960 > 231 = b (# of blocks of the design)
10/12
JI
Back
FullScr
Theorem. There are three quasi-symmetric (56, 16, 18) designs withx = 4, y = 8 and G = 〈α, β〉 as an automorphism group. The firstone has G as its full automorphism group, the second one has a splitextension G.Z2 of order 1920, and the third one has a split extensionM21.Z2 of order 40320 as its full automorphism group.
Computational details:
• total number of orbits of 16-element subsets is >(56
16)960 ≈ 4.34 · 1010
• most orbits are of size 960 > 231 = b (# of blocks of the design)
• there are 1242 “short orbits”, of size ≤ b; we can generate themdirectly by an algorithm written in GAP
10/12
JI
Back
FullScr
Theorem. There are three quasi-symmetric (56, 16, 18) designs withx = 4, y = 8 and G = 〈α, β〉 as an automorphism group. The firstone has G as its full automorphism group, the second one has a splitextension G.Z2 of order 1920, and the third one has a split extensionM21.Z2 of order 40320 as its full automorphism group.
Computational details:
• total number of orbits of 16-element subsets is >(56
16)960 ≈ 4.34 · 1010
• most orbits are of size 960 > 231 = b (# of blocks of the design)
• there are 1242 “short orbits”, of size ≤ b; we can generate themdirectly by an algorithm written in GAP
• only 40 of the short orbits are good, with intersection numbersx = 4, y = 8
10/12
JI
Back
FullScr
Theorem. There are three quasi-symmetric (56, 16, 18) designs withx = 4, y = 8 and G = 〈α, β〉 as an automorphism group. The firstone has G as its full automorphism group, the second one has a splitextension G.Z2 of order 1920, and the third one has a split extensionM21.Z2 of order 40320 as its full automorphism group.
Computational details:
• total number of orbits of 16-element subsets is >(56
16)960 ≈ 4.34 · 1010
• most orbits are of size 960 > 231 = b (# of blocks of the design)
• there are 1242 “short orbits”, of size ≤ b; we can generate themdirectly by an algorithm written in GAP
• only 40 of the short orbits are good, with intersection numbersx = 4, y = 8
• the 7 × 40 Kramer-Mesner system has 5 solutions respecting thecompatibility matrix, giving rise to the 3 quasi-symmetric designsdescribed in the theorem
11/12
JI
Back
FullScr
The binary linear codes generated by the block incidence vectors of thenew designs are self-orthogonal (C⊥ ≤ C).
11/12
JI
Back
FullScr
The binary linear codes generated by the block incidence vectors of thenew designs are self-orthogonal (C⊥ ≤ C).
The first and the second design span codes of length 56, dimension 23and minimum distance 8. Best known [56, 23, d ]2 code: d = 14.
11/12
JI
Back
FullScr
The binary linear codes generated by the block incidence vectors of thenew designs are self-orthogonal (C⊥ ≤ C).
The first and the second design span codes of length 56, dimension 23and minimum distance 8. Best known [56, 23, d ]2 code: d = 14.
The third design (with automorphism group M21.Z2) spans a code oflength 56, dimension 19 and minimum distance 16. Same parametersas the best known [56, 19, d ]2 code!
11/12
JI
Back
FullScr
The binary linear codes generated by the block incidence vectors of thenew designs are self-orthogonal (C⊥ ≤ C).
The first and the second design span codes of length 56, dimension 23and minimum distance 8. Best known [56, 23, d ]2 code: d = 14.
The third design (with automorphism group M21.Z2) spans a code oflength 56, dimension 19 and minimum distance 16. Same parametersas the best known [56, 19, d ]2 code!
Description from M. Grassl’s page www.codetables.de based onA. E. Brouwer’s tables:
Construction of a linear code [56, 19, 16] over GF (2):
1: [55, 21, 15] Cyclic Linear Code over GF (2). CyclicCode of length 55with generating polynomial x34 +x31 +x29 +x28 +x26 +x23 +x19 +x18 + x13 + x10 + x7 + x5 + x3 + x + 1.
2: [56, 21, 16] Linear Code over GF (2). ExtendCode [1] by 1.
3: [56, 19, 16] Linear Code over GF (2). Subcode of [2].
12/12
JI
Back
FullScr
Is the code generated by the third (56, 16, 18) design equivalent to aknown [56, 19, 16]2 code?
12/12
JI
Back
FullScr
Is the code generated by the third (56, 16, 18) design equivalent to aknown [56, 19, 16]2 code?
Problem: number of subcodes in step 3 is
[21
19
]2
= 733 006 703 275.
12/12
JI
Back
FullScr
Is the code generated by the third (56, 16, 18) design equivalent to aknown [56, 19, 16]2 code?
Problem: number of subcodes in step 3 is
[21
19
]2
= 733 006 703 275.
Idea: look at minimum weight words in the [56, 21, 16]2 code from step 2.There are 5170 words of weigt 16; quasi-symmetric designs are equiva-lent to b = 231 words pairwise intersecting in x = 4 or y = 8 coordi-nates.
Computational tools: GAP package GUAVA
cliquer by P. Ostergard and S. Niskanen
12/12
JI
Back
FullScr
Is the code generated by the third (56, 16, 18) design equivalent to aknown [56, 19, 16]2 code?
Problem: number of subcodes in step 3 is
[21
19
]2
= 733 006 703 275.
Idea: look at minimum weight words in the [56, 21, 16]2 code from step 2.There are 5170 words of weigt 16; quasi-symmetric designs are equiva-lent to b = 231 words pairwise intersecting in x = 4 or y = 8 coordi-nates.
Computational tools: GAP package GUAVA
cliquer by P. Ostergard and S. Niskanen
Thanks for your attention!